1. Field of the Invention
The present invention generally relates to methods and systems for detection of defects particularly in relatively noisy inspection data Certain embodiments relate to a computer-implemented method for detecting one or more selected types of defects in relatively noisy inspection data based on spatial characteristics of the defects.
2. Description of the Related Art
Fabricating semiconductor devices such as logic and memory devices typically includes processing a substrate such as a semiconductor wafer using a number of semiconductor fabrication processes to form various features and multiple levels of the semiconductor devices. For example, lithography is a semiconductor fabrication process that involves transferring a pattern from a reticle to a resist arranged on a semiconductor wafer. Additional examples of semiconductor fabrication processes include, but are not limited to, chemical-mechanical polishing, etch, deposition, and ion implantation. Multiple semiconductor devices may be fabricated in an arrangement on a semiconductor wafer and then separated into individual semiconductor devices.
As the dimensions of advanced semiconductor devices continue to shrink, the presence of defects in the semiconductor devices increasingly limits the successful fabrication, or yield, of the semiconductor devices. For example, a scratch formed on a wafer during chemical-mechanical polishing may cause an open circuit or a short circuit in, or complete failure of, one or more semiconductor devices formed in subsequent processing. Because fabrication of a semiconductor device includes many complex process steps, the adverse effects of defects on total yield may increase exponentially if a defect formed on a wafer in one manufacturing process step causes additional defects to be formed on the wafer in subsequent manufacturing process steps.
Accordingly, defect detection or “inspection” of semiconductor wafers is and will continue to be of significant importance in semiconductor development and manufacturing. Furthermore, the ability of inspection tools or systems to detect a range of defect types over a range of sensitivities will determine how well defects can be detected and, therefore, how well semiconductor fabrication processes can be monitored and controlled. Consequently, significant efforts have been and will continue to be made to increase the sensitivity of inspection systems by improving parameters of the systems such as resolution. There have also been significant efforts in improving the processing of inspection data to increase the accuracy with which defects can be detected.
However, most inspection data processing involves two steps: defect detection and then classification. For instance, on many commercially available inspection systems, defects are found by detecting point defects via signal thresholding on individual data points in simple one-dimensional scans. Individual point defects may then be displayed on a point defect map or organized into another format. The point defect map is then post-processed to recognize if several of the points fall roughly into a two-dimensional shape, at which point that collection of points is labeled or classified as a specific defect instead of as individual particle defects.
There are, however, several disadvantages to the above methods of inspection data processing for detecting the presence of particular types of defects. In particular, these methods can be relatively inaccurate when detecting defects in relatively noisy inspection data. For example, as described above, simple one-dimensional scans only generate raw signals at individual points on the substrate, and every encounter with a two-dimensional surface anomaly is treated as a disconnected collection of point defects. Therefore, signal thresholding yields a defect map determined solely by signal strength at the individual points. Consequently, portions of faint two-dimensional defects may be lost to background noise due to failure of some of its associated point defect signals to exceed the threshold. As a result, the above-described methods for detecting defects may be substantially inaccurate when detecting defects in noisy inspection data since many defects may not be detected at all. In addition, the above-described methods for detecting defects may be substantially inaccurate in detecting the types of defects that are present on a substrate since portions of defects may not be detected thereby increasing the probability of misclassification of defects.
Accordingly, it may be advantageous to develop methods and systems for detecting defects on a substrate that are substantially accurate for detecting a range of defect types having a range of sensitivities, particularly in relatively noisy inspection data.